A machine learning-based tool for predicting outcomes of stellar encounters using neural networks trained on SPH (Smoothed Particle Hydrodynamics) simulations.

Installation

Requirements

pip install torch numpy h5py

Overview

This model classifies stellar encounters into three physical regimes and predicts collision outcomes:

Encounter Regimes

• Collision (flag: -1): Physical contact between stars when pericenter < R₁ + R₂
• Tidal Capture (flag: -2): Stars become gravitationally bound through tidal energy dissipation
• Flyby (flag: -3): Stars pass without significant interaction

Output Format

Each result is a dictionary containing:

• regime_flag: Integer indicating the encounter type
  • -1: Collision
  • -2: Tidal capture
  • -3: Flyby
• predicted_class: Classification outcome (0-3 for collisions, 1 for tidal capture, 2 for flyby)
• predicted_values: List of [star_mass1, star_mass2, unbound_mass] in M☉

Tutorial

• See ./examples/Tutorial.ipynb for detailed examples and workflow demonstrations.
• See ./examples/NN_tutorial.ipynb for detailed example on how to use the NNs independently.
• See ./examples/MoE_tutorial.ipynb for detailed examples on how to use the MoE independently.

Key Assumptions & Caveats

⚠️ Stellar Evolution

• Main Sequence Only: Stars must be on the main sequence (MS). Post-TAMS (Terminal Age Main Sequence) stars will raise a ValueError
• Metallicity: Model uses tracks at Z = 0.01 Z☉ (solar metallicity)
• Stellar radii are interpolated from POSYDON grids based on age and mass

⚠️ Mass Range

• Validated for masses between ~0.1 - 100 M☉
• Extrapolation outside this range may be unreliable

⚠️ Tidal Dissipation Physics

Uses polytrope approximations from Portegies Zwart & McMillan (1993) and Mardling & Aarseth (2001):

• n = 1.5 for M < 0.8 M☉ (convective envelopes)
• n = 3.0 for M > 0.8 M☉ (radiative envelopes)
• Linear interpolation for 0.4 < M < 0.8 M☉ (mixed structure)

⚠️ Simplified Regime Assumptions

• Tidal Capture: Assumes perfect merger with no mass loss
• Flyby: Assumes no mass transfer or interaction
• All regimes: No stellar rotation or stellar winds considered

Error Handling

The code will raise errors for:

• Post-TAMS stars: ValueError when central H fraction < 10⁻⁵
• Mismatched array lengths: All input arrays must have same length
• Missing model files: Check that .pt files are in correct location

References

• Portegies Zwart, S. F., & McMillan, S. L. W. (1993). The evolution of close triple stars. ApJ, 410, 759
• Mardling, R. A., & Aarseth, S. J. (2001). Tidal interactions in star cluster simulations. MNRAS, 321, 398
• Fragos, T., et al. (2023). POSYDON: A Population Synthesis Code
• MESA stellar evolution code (Paxton et al. 2011, 2013, 2015, 2018, 2019)

Citation

If you use this code in your research, please cite:

González Prieto, E., et al., 2026, arXiv:2602.10191

License

[Add your license here]

Contact

elena.prieto[at]northwestern.edu

Acknowledgments

This work uses:

• POSYDON v2 stellar evolution grids
• PyTorch for neural network implementation
• SPH simulation data for model training